Sunteți pe pagina 1din 12

PHYSICAL REVIEW E 86, 051305 (2012)

Stratification, segregation, and mixing of granular materials

in quasi-two-dimensional bounded heaps

Yi Fan,1 Youcef Boukerkour,2 Thibault Blanc,2 Paul B. Umbanhowar,1 Julio M. Ottino,1,3,4 and Richard M. Lueptow1,*
Department of Mechanical Engineering, Northwestern University, Evanston, Illinois 60208, USA
French Air Force Academy, Salon de Provence, France
Department of Chemical and Biological Engineering, Northwestern University, Evanston, Illinois 60208, USA
The Northwestern University Institute on Complex Systems (NICO), Northwestern University, Evanston, Illinois 60208, USA
(Received 29 August 2012; published 29 November 2012)
Segregation and mixing of granular mixtures during heap formation has important consequences in industry
and agriculture. This research investigates three different final particle configurations of bidisperse granular
mixturesstratified, segregated and mixedduring filling of quasi-two-dimensional silos. We consider a large
number and wide range of control parameters, including particle size ratio, flow rate, system size, and heap rise
velocity. The boundary between stratified and unstratified states is primarily controlled by the two-dimensional
flow rate, with the critical flow rate for the transition depending weakly on particle size ratio and flowing layer
length. In contrast, the transition from segregated to mixed states is controlled by the rise velocity of the heap, a
control parameter not previously considered. The critical rise velocity for the transition depends strongly on the
particle size ratio.

DOI: 10.1103/PhysRevE.86.051305 PACS number(s): 45.70.Mg, 81.05.Rm, 47.57.Gc

I. INTRODUCTION velocity ratio of different sublayers) can only be determined

by fitting experimental or simulation data, limiting the general
Heap flow of granular materials occurs in many contexts
applicability of this model.
[1,2]. For example, when granular materials such as powders,
Stratification in heap flow was studied in detail by Makse
grains, or pelletized polymers flow into the top of a container,
et al. [11,13]1 and by Gray and colleagues [12,19]. In
a heap builds, where the granular material tumbles down the
experiments, they found that when the components of granular
pile in a flowing layer that is on the order of ten particle
mixtures differ in both size and shape, mixtures of larger
diameters thick. When materials are mixtures of particles
rough and smaller smooth particles stratify into alternating
differing in size, density, shape, and/or surface properties,
layers of each particle type. Gray and colleagues [12,19] pro-
different components tend to distribute inhomogeneously.
posed a mechanism involving surface avalanches and upslope
In some cases, larger particles flow further down the heap
propagating shock waves. Makse et al. [11,13] attributed the
than smaller particles, resulting in segregation [310]. In
formation of stratified layers to competition between size seg-
other cases, large and small particles form alternating layers,
regation and shape segregation, and proposed that stratification
resulting in stratification [1120]. In still other situations, large
only occurs for mixtures of large rough and small smooth
and small particles remain mixed [16].
particles. They also adapted a continuum model [13,21,22]
In what were perhaps the earliest attempts to understand
and a cellular automaton model [11,13,22] to successfully
the physical mechanisms driving segregation in heap flow,
reproduce the stratification observed in their experiments.
Williams [3,4] and Drahun and Bridgwater [5] performed
However, Baxter et al. [16] found that stratification can also
heap-flow experiments using bidisperse mixtures of different-
occur for different-sized smooth spherical particles, though
sized spherical particles. They proposed a percolation mech-
they did not perform a systematic study over a wide range of
anism for heap segregation in which small particles tend to
parameters. Further, they also mentioned that a mixed state (no
sink through voids preferentially, while large particles rise
segregation or stratification) exists in certain situations such
to the free surface and roll to the end of the flowing layer.
as at high feed flow rates.
Consequently, small particles accumulate below the upstream
Although three final states of heap flow of bidisperse
portion of the flowing layer at the center of the heap, while large
granular materials (segregated, stratified, and mixed) have
particles accumulate at the downstream end of the flowing
been observed and studied by different researchers, none of
layer adjacent to the bounding outer walls. Based on this
the past research appears to have investigated the dependence
picture, Shinohara and co-workers [6,7] developed a screening
of the final particle distributions on a broad range of control
layer model based on conservation equations incorporating
parameters (see Table I), including volumetric feed rate Q,
the percolation mechanism. A recent experimental and com-
silo width W , 2D silo gap thickness T , species size ratio
putational study [10] performed to test the screening layer
R = Dl /Ds (where Dl and Ds are the large and small
model [7] showed that the model captures some key features of
particle diameter, respectively), and absolute particle size. The
heap segregation in certain ranges of experimental parameters.
focus of this research is to systemically explore how these
However, some variables in the screening layer model (such
as the penetration rate of segregating components and the

One figure of Williams [4] shows stratification in heap flow, but
* the stratification was not noted.

1539-3755/2012/86(5)/051305(12) 051305-1 2012 American Physical Society

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

TABLE I. Experimental parameters in past and current research.

Size ratio Silo width 2D silo thickness Flow rate Heap stage
References Segregation type R W (cm) T (cm) Q (cm3 /s) (see Fig. 1)

Williams [3] Size 5.2 31a 2.5 Unspecified II

Drahun & Bridgwater [5] Size 1.32.0 43 Unspecified Unspecified III
Shinohara et al. [6] Size 14.3 30 Unspecified 3080 II&III
Shinohara & Enstad [7] Size 2.015.0 1518 Unspecified 1433b III
Baxter et al. [16] Size 2.0 50a 9 6, 736c II
Thomas [8] Size 2.050.0 10 3D silo 2740c II
Goyal & Tomassone [9] Size 1.35.0 22 0.5 Unspecified II
Rahman et al. [10] Size 9.215.2 1518 3D silo 33b III
Makse et al. [11,13] Size & shape 1.76.7 30 0.51 Unspecified II
Gray et al. [12,19] Size & shape 1.471.7 37 0.3 Unspecified II&III
Grasselli & Herrmann [14] Size & shape 1.210.5 30 0.10.6 0.23.5c II
Koeppe et al. [15] Size & shape 2.0 27 0.32.4 0.57.4c II
Shimokawa & Ohta [17,18] Size & shape 2.08.0 60 0.5 0.081.72d II
Current study Size 1.36.0 2291 0.62.5 1420 III
For particles fed at the silo center, W is half of the silo width.
Flow rate is estimated as Q = F /0.6, where F is feed flow rate based on net particle volume.
Flow rate is estimated as Q = Qm /(0.6m ), where Qm is mass flow rate and m is material density.
Q = Qm /(0.6m ), where Qm is mass flow rate and m is estimated as 2.0 g/cm3 .

parameters affect transitions between different final particle heap becomes angled in stage II and grows laterally until
configurations. it reaches the bounding end wall. In stage III, the laterally
Figure 1(a) is a sketch of the top view of a three-dimensional constrained heap rises steadily at a constant rise velocity vr .
(3D) silo, which typically comprises a vertical cylindrical We examine the steady filling stage (III), where the length of
container where granular material falls vertically along its the flowing layer L is constant.
centerline and flows radially in all directions down the heap, In this article, we present an experimental study of heap
filling the container to its outer wall. Here, we use a quasi-two- segregation of granular mixtures of binary spherical particles
dimensional (2D) silo, which can be thought of as a section of differing only in size in a quasi-2D silo for a larger number and
the 3D silo as shown in Fig. 1(a). This geometry makes it easy wide range of experimental parameters than in previous work
to observe the final particle distributions and minimizes the (Table I). We observed all three final particle configurations
volume of particles needed. In the quasi-2D silo, rather than stratified, segregated, and mixed. We find that for constant R,
feed rate Q, the relevant flow rate is the 2D volumetric flow rate the transition from the stratified state to the unstratified state
down the slope at the peak of the heap, defined as q = Q/T , is controlled by q, while the transition from the segregated
which decreases linearly along the flow direction. As shown state to the mixed state is controlled by the heap rise velocity
in Fig. 1(b), filling of a silo proceeds in three stages. In stage I, vr = q/W . Both of these transitions depend on the size ratio
a somewhat irregularly shaped initial heap forms. Shortly, the R. Phase diagrams are presented that illustrate the effects of R,
W , and q or vr on transitions between different final particle
configurations. We further provide insight into the stratification
(a) (b) Q of mixtures of spherical particles with equal repose angles, in
contrast to previous studies [11,12] that used large rough and
vr = Q/A
small smooth particles with different repose angles. Finally,
we propose a dimensionless velocity ratiothe ratio of rise
W ste velocity to particle percolation velocityas a key control
T ady
fillin parameter for the segregated state.
late g (I
ral II) In the remainder of this paper, Sec. II describes the experi-
init wth mental setup. Sections III and IV present experimental results
iati (II)
(I) and discussion thereof. Section V presents our conclusions.

FIG. 1. (Color online) (a) Top view sketch of a 3D silo, where
arrows indicate flow of material. Dashed box shows top view of the The quasi-2D silo in our experiments consists of a pair
quasi-2D silo used here. (b) Side view sketch of a quasi-2D silo rising of 91 69 1.27 cm vertical rectangular plates: one is glass
at the rise velocity vr = Q/A, where Q is the volumetric feed rate and for observation and measurement purposes and the other is
A = T W . The three stages of heap flow are: (I) initiation, (II) lateral aluminum to reduce electrostatic charging. Vertical spacer bars
growth, and (III) steady filling (downstream end of heap constrained were clamped between the parallel plates to control the silo gap
by outer wall). thickness T and to vary the silo width W . Granular mixtures


TABLE II. Binary mixtures of glass particles with equal mass TABLE III. Parameters.
Parameter Description Type
Size ratio Large particles Small particles
R Dl (mm)a Ds (mm)a Q 3D flow rate Controlled
W Silo width Controlled
1.3 2.00 0.08 1.51 0.09 T Silo gap thickness Controlled
1.5 1.69 0.05 1.14 0.08 R = Dl /Ds Size ratio Controlled
2.0 2.00 0.08 1.00 0.06 q = Q/T 2D flow rate Derived
2.2 1.10 0.06 0.50 0.05 vr = q/W Heap rise velocity Derived
3.4 1.69 0.05 0.50 0.05
4.0 2.00 0.08 0.50 0.05
6.0 2.98 0.05 0.50 0.05 systematically varying control parameters (Table III) including
a the feed volumetric flow rate Q, system size (silo width W and
Mean particle diameter with standard deviation.
silo thickness T ), size ratio R, and absolute particle size.

were fed either at one end of the silo or at the centerline of the
silo (the length of the flowing layer L in the latter case is half
of the former case). Similar results were observed independent Stratified, segregated, and mixed final states noted by
of whether the feed was at one end of the silo (total silo width previous researchers were all observed in our experiments for
W ) or at the centerline of the silo (total silo width 2W ). For the stage III, as shown in Fig. 2. The thin free surface layer should
results presented here, values for W were 22, 46, 69, and 91 cm. be ignored, as it is associated with residual flow at the end
We note that the size of the silo at W = 91 cm is comparable of filling. At small flow rates, similar but more pronounced
to small, full-scale, industrial silos. The effect of the silo gap stratification than reported in Baxter et al. [16] occurs [see
thickness T on the transition between segregation and mixing Fig. 2(a)]. The stratified state consists of alternating layers
is negligible, provided that T is more than four large particle of small and large particles parallel to the flow direction,
diameters. However, stratification depends sensitively on T , coexisting with segregation along the flow direction, where the
as discussed in the Appendix. Here, we use T = 1.27 cm for downstream region of the heap contains mostly large particles
all experiments, unless otherwise noted. and the upstream region of the heap close to the feed point
Seven combinations of different-sized soda-lime glass contains mostly small particles. Stratification becomes weaker
particles with size ratios R ranging from 1.3 to 6.0 were and eventually disappears as q is increased toward a critical
investigated (see Table II). We limit R to be less than (2/ 3 value (to be discussed shortly). Above this critical value full
1)1 = 6.46, above which spontaneous percolation occurs segregation is observed, as shown in Fig. 2(b). In this regime,
[2325] (see Sec. IV C). To distinguish different species, the heap consists of two distinct regions: The downstream
different colors of particles were used. To ensure roundness region consists of nearly all large particles, while the upstream
and similar surface properties such as friction coefficient region consists of a few large particles scattered in a sea of
between different species, we purchased particles colored by small particles. The boundary between these two segregated
the manufacturer (Sigmund Lindner GmbH, Germany). The regions is narrow. As q is further increased, the region of larger
granular mixtures in our experiments were composed of either particles in the downstream portion of the heap shrinks and
metal-coated, surface-colored red and blue particles, or clear more large particles remain in the upstream portion of the heap.
and surface-colored black particles. Several trial experiments One might expect that at high enough q, the region containing
showed that final particle configurations are insensitive to the all large particles disappears and a perfectly mixed state is
surface coatings of the two components. The material density achieved everywhere in the silo. However, due to limitations
of the particles is 2.59 g/cm3 . of our experimental apparatus, the silo cannot always be filled
A small auger feeder (Acrison, Inc., NJ, USA) dispensed at a high enough flow rate q to achieve perfect mixing. Instead,
the granular mixtures into the silo. The auger feeder produced for most experiments at the highest achievable q, a near mixed
a stable and reproducible flow rate over a wide range of state [see Fig. 2(c)] is obtained where only a narrow large
volumetric flow rates (1420 cm3 /s). Flow rate was varied particle region exists at the downstream end of the heap and
by controlling the rotation frequency of the motor and the the remainder of the heap is well mixed.
diameter of the auger. The granular mixtures were composed
of equal masses of large and small particles and were well
mixed upon filling the auger feeder. We performed several test A. Stratification
experiments to measure the mass fraction of the two species of Stratification [see Fig. 2(a)] occurs at small flow rates
different-sized particles after discharge from the auger feeder over a wide range of size ratios and silo widths. We use the
at several different flow rates and size ratios. We found that the image intensity to quantify the final particle distributions for
mass fraction of the two components remained at 50:50 (within stratification for each experimental run. Since the particles
3%) during the entire experiment. Before each experiment, used in the experiments have different colors, the local
antistatic spray (Sprayon, OH, USA) was applied to the glass image intensity I is monotonically related to local particle
wall to limit electrostatic effects. A digital camera in front concentration. In all our experiments, small particles have
of the glass wall recorded the filling and the final state of higher intensity than large particles, so higher local intensity
the heap. We performed more than 400 experimental runs, implies higher local concentration of small particles.

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

10 cm (a) (b) (c)

FIG. 2. (Color online) Three final particle configurations of bidisperse granular mixtures in quasi-2D heap flow for R = 3.4, T = 1.27 cm,
and W = 69 cm (the free surface layer is associated with residual flow at the end of filling and should be ignored). (a) Stratification at
q = 0.8 cm2 /s; (b) segregation at q = 18.9 cm2 /s; and (c) nearly complete mixing at q = 328 cm2 /s. Dark (blue online): 1.69-mm glass
spheres; light (red online): 0.5-mm glass spheres. The width of each image is W .

Figure 3 illustrates our method for measuring particle y [Fig. 3(c)]. The intensity profile represents the variation
concentration for stratification. We study the region outlined of species concentration due to stratification; the periodic
in Fig. 3(a), which is located in stage III of the heap formation intensity oscillation in the y direction corresponds to the
and excludes the free surface region associated with residual alternating layers of small and large particles.
flow at the end of filling. The outlined region is transformed To further quantify the stratification globally and to de-
to a rectangular box [see Fig. 3(b)] by rotating by the angle of termine the transition to segregation at different experimental
repose in the counterclockwise direction and then shearing conditions, we calculate the standard deviation of the intensity
in the horizontal direction. The image intensity is averaged N in Fig. 3(c) over the range of y coordinates, I =
in the x direction (0  x  L) and plotted as a function of i=1 (Ii I ) /(N 1), where I is the mean intensity, Ii

is the intensity at row i, and N is the number of pixel rows

in the y direction. Larger I /I indicates a higher degree of
(a) stratification. When there is no stratification, I /I goes to a
constant residual value of 0.005 associated with variations in
lighting intensity and random fluctuations in concentration.
Figure 4 shows I /I as a function of q for R = 2.2.
I /I is significantly larger at small q corresponding to strong
stratification (long layers). As q increases, I /I decreases as
the stratified layers become shorter and stratification weakens.
I /I decreases to a small constant value at a transitional
2D flow rate qt , where stratification disappears and only
segregation occurs. The transition from a stratified state to
an unstratified state occurs around qt = 6 cm2 /s, independent
x W = 91 cm
0 W = 69 cm
W = 46 cm

y (cm)


90 100 110 120 130
I (arbitrary units)
FIG. 3. (Color online) Image processing method for quantifying
stratification. (a) Image from the experiment with W = 91 cm, T = FIG. 4. (Color online) I /I showing decrease in stratification
1.27 cm, and q = 1.8 cm2 /s showing a stratified mixture of 0.5-mm with increasing q for R = 2.2 at different W . Insets: Images from
(light) and 1.1-mm (black) glass particles. (b) Dashed parallelogram experiments at the indicated data points, where Ds = 0.5 mm for
region from (a) transformed to a rectangular box. (c) Intensity I from small light particles, Dl = 1.1 mm for large black particles, and W =
(b) averaged over x and plotted as a function of y. 91 cm.



5 stratified segregated mixed


1 (c)
0 1 2 3
10 10 2
10 10 200
q (cm /s)

I (arbitrary units)
FIG. 5. (Color online) Phase diagram of final states (stratified:
red circle; segregated: black square; mixed: blue diamond) in terms
of q and R at three different W . Data are artificially offset in R to 100
show each data point for different W : from top to bottom, W = 91,
69, and 46 cm, respectively. Horizontal solid lines denote the actual 50
size ratio to guide the eye. Dashed line segments mark the boundary
between stratified and unstratified states.
0 0.5 1
of W . Similar trends are observed for all other R except the x/L
smallest value considered, R = 1.3, where no stratification
occurs for all q and W tested. FIG. 6. (Color online) Image processing method for quantifying
The influence of q, W , and R on the transition from segregation. (a) An image from the experiment with W = 69 cm,
stratified to unstratified states is examined by plotting a phase T = 1.27 cm, and q = 10.9 cm2 /s for a mixture of 0.5-mm (light)
diagram as a function of q and R, shown in Fig. 5. At each and 1.1-mm (black) glass particles. (b) The region in the dashed
R, data for different values of W are artificially offset in three parallelogram from (a) transformed into a rectangle. (c) Intensity I
rows representing from top to bottom, W = 91 cm, W = 69 from (b) averaged over y and plotted as a function of x/L. L denotes
cm, and W = 46 cm, respectively, to show all of the data the width of the dark region of large particles at the end of the heap,
and L is the length of the flowing layer.
points. The three final states are distinguished by symbols and
the dashed line indicates the boundary between the stratified
state and the unstratified state. (We discuss the transition from parallelogram is transformed to a rectangle as in Fig. 6(b), and
segregation to mixing in Sec. III B.) the image intensity is averaged in the y direction and plotted
For 2  R  4, stratification occurs when q is less than as a function of x as shown in Fig. 6(c) for a typical case.
6 cm2 /s. At these size ratios, the transition from the stratified The upstream portion of the heap (x/L  0.6) has a smoothly
state to the unstratified state occurs at the same transitional varying concentration of particles of the two different sizes,
flow rate qt for all three values of W . For R = 6, stratification whereas the downstream end (x/L > 0.6) has a nearly uniform
is observed up to q = 12 cm2 /s for all three values of W . concentration of only large particles.
This increase in qt is possibly because R is close to the Figure 7(a) shows a series of intensity profiles plotted as a
size ratio for spontaneous segregation, 6.464, so that other function of x/L at different W and q at R = 2.2. The profiles
factors such as wall effects or horizontal segregation due to overlay one another for different values of W , as discussed
spanwise shear rate gradients [26] may significantly influence shortly. Each profile represents the final state distribution of the
the stratification. two segregated species. The concentration of small particles
At small R, stratification diminishes. When R = 1.3, no is higher in the upstream region and the concentration of large
stratification occurs for all W . When R = 1.5, stratification particles is higher in the downstream region. The boundary
occurs only at W = 69 and 91 cm; no stratification is observed between these two regions is narrow (less than 0.2L). Close
at W = 46 cm. Thus, W may also affect the occurrence of the to the feed zone of the heap (x/L  0.2), the concentration
stratification at small R, as discussed further in Sec. IV A. of small particles is slightly smaller than in the rest of the
upstream region [such as in Fig. 6(c) and Fig. 7(a)]. This
likely occurs because the incoming particles discharged from
B. Segregation the auger feeder start flowing from a nearly stationary state
Full segregation occurs when q increases beyond the after falling onto the heap so that the segregation is weaker in
transitional value qt . A similar image processing method to this region.
that for stratification is used to quantify segregation. As shown As indicated in Fig. 7(a), the similar intensity distributions
in Fig. 6(a), the outlined region in stage III is considered. The in each plot correspond to identical rise velocities vr but

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

(a) 200
W = 46 cm 6
W = 69 cm
100 vr = 0.08 cm/s W = 91 cm

0 5 stratified segregated mixed

I (arbitrary units)


100 4
vr = 0.24 cm/s

vr = 1.2 cm/s
0 0.2 0.4 0.6 0.8 1
(b) 0.5 10 10
10 10
W = 46 cm
W = 69 cm vr (cm/s)
0.4 W = 91 cm
FIG. 8. (Color online) Phase diagram of final state (stratified:
red circle; segregated: black square; mixed: blue diamond) in terms
of rise velocity vr and size ratio R at three silo widths W . Data
L/ L

are artificially offset in R to show each data point for different W :

0.2 from top to bottom, W = 91, 69, and 46 cm, respectively. Horizontal
solid lines denote the actual size ratio to guide the eye. Dashed line
0.1 segments indicate the boundary between segregation and mixing.

To determine the transition from segregation to mixing as

0 a function of vr and R, we quantify the degree of segregation
0 1 2 3 4 5
vr (cm/s) using the normalized length of the large particle region L/L.
This measure is similar to the approach of Shinohara et al.
FIG. 7. (Color online) (a) Intensity I as a function of x/L for [6], who used 1 L/L. In stage III, L/L represents the
a mixture of 0.5-mm (light) and 1.1-mm (black) particles at vr = approximate mass of segregated large particles relative to the
0.08 cm/s (top), 0.24 cm/s (middle), and 1.2 cm/s (bottom). (b) mass of the mixture. Therefore, based on mass conservation,
L/L vs vr for the same mixture at different silo widths W . Insets 0  L/L  0.5, where L/L = 0 means perfect mixing
show images for W = 69 cm at indicated data points. Dashed line is and L/L = 0.5 means complete segregation.
the fit L/L = (vr /a)b , where a =0.017 cm/s and b = 0.493. Figure 7(b) plots L/L as a function of vr for different
W at R = 2.2. At other size ratios similar trends occur. As
different silo widths W and flow rates q. In stage III, vr = vr is increased, L/L decreases, indicating that segregation
Q/(W T ) = q/W and is independent of the angle of repose becomes weaker at higher vr . For vr  1.5 cm/s, L/L <
that changes slightly as q or R varies [27]. Thus, when W and 0.1, which means more than 80% of the large particles are
q are varied together, vr , instead of q, controls the final particle mixed with small particles in the upstream region of the
distributions for segregation. In other words, for a given heap (x/L < 0.9). This corresponds to a mass ratio of small
mixture at the same vr , small and large particles distribute particles in this region of 56%. Curves of L/L as a function
similarly along the flow direction for different values of W and of vr at different W collapse onto a single curve, consistent
q. When vr increases, more large particles stay in the upstream with the collapse shown in Fig. 7(a) for different W .
region so that the concentration of small particles decreases Due to limitations of our experimental apparatus (particles
and the width of the downstream region of large particles overflowing the entrance of the silo), a rise velocity greater than
L decreases as well, as shown in Fig. 7(a). However, the 10 cm/s (equivalent to filling the entire silo in less than 6 s)
concentration of large particles in the downstream region of cannot be achieved. A perfect mixed state is therefore difficult
the heap does not change as vr increases (the intensity at large to obtain at most R except for R = 1.3, where relatively good
x/L is constant at different vr ). Based on our experiments, mixing is observed for vr > 1 cm/s. We fit the experimental
there is almost always a region at the end of the heap close data in Fig. 7(b) to a power law as L/L = (vr /a)b using
to the bounding walls containing nearly all large particles, the least-squares method for each size ratio [see Fig. 7(b) for
regardless of the flow rate, silo width, and size ratio.2 R = 2.2], where the fitting parameters a and b depend on R.
Based on this fit, a cutoff value for L/L = 0.15 is selected

There are two exceptions: (1) when R < 1.5, the large particle
region at the end of the heap is absent in many cases, which is the bouncing of particles. For all data presented in this paper, the
discussed in Sec. IV C. (2) At large Ds or small W , a narrow region of bouncing-induced segregation is minimized by using small particles
small particles appears at the end of heap due to another mechanism: and large W (see Sec. IV D).


to determine the transitional rise velocity between segregation 30

and mixing, so that the boundary can be determined at different
R. This value corresponds to a small particle mass ratio of 59% T=1.27 cm

Angle of repose (degree)

in the region x/L < 0.85.
T=1.91 cm
Using L/L = 0.15 as the cutoff value, a phase diagram
similar to that in Fig. 5, but in terms of vr instead of q, T=2.54 cm
is constructed to quantify the effects of R, W , and vr on
the transition from segregation to mixing in Fig. 8. The
dashed line segments in the figure mark the approximate
boundary between segregation and mixing. The phase diagram
demonstrates that segregation transitions to mixing at the same
vr for different W at the same R. vr at the transition increases
by roughly 1 order of magnitude as R is increased from 1.3 to
6.0. 20
0 1 2 3
Particle diameter (mm)
FIG. 9. (Color online) Static angle of repose for monodisperse
As shown in Sec. III, stratification generally occurs at low glass particles at different silo gap thicknesses. Results are for 0.5,
flow rates, and the transition between stratified and unstratified 1.1, 1.7, 2, and 3 mm monodisperse particles flowing into the silo
states depends on the flow rate q. In contrast, the transition at identical feed rates. The static angle of repose is measured after
between segregated and mixed states depends on the rise filling is stopped.
velocity of the heap vr . In this section, we discuss possible
mechanisms for these phenomena. Further, we briefly discuss, appearance of layers of spherical particles of different sizes
with respect to segregation, the limits of size ratio and the is associated with the formation of a series of kinks on the
effect of particle bouncing after initial impact with the heap. slope of the heap away from the boundaries during intermittent
avalanches that occur only at low flow rates. As Fig. 10 shows,
A. Stratification dynamics during an avalanche, large (dark) particles segregate to the
As mentioned in Sec. I and reported in many studies free surface and roll down the free surface forming the front of
[1115,1719] and references therein, stratification has been an avalanche. This front of large particles ceases to flow and
most often observed in mixtures of large rough and small
smooth particles. The stratification of large rough and small 10 cm (a)
smooth particles is usually attributed to differences in the angle
of repose: steeper for rough particles than for smooth spherical
particles. This difference in repose angle induces a kink at the kink A
base or the bounding wall of the silo [1115,19] or on the slope
of the heap [17,18], with a stratified layer upstream of the kink.
The stratified layer is then frozen by an upslope propagating
shock wave [12,19,20] and the process repeats. Although
Williams [3] and Baxter et al. [16] showed some evidence
of stratification for different-sized smooth spherical particles,
the existence of stratification in mixtures of different-sized
spherical particles has been debated,3 and only a few limited
examples of stratification have been provided [3,16]. (b)
In this research, all particles are spherical so shape effects
are excluded. Furthermore, except for the smallest particles, kink B
the angle of repose is nearly independent of particle size
at the same silo gap thickness T , similar to Goyal and
Tomassone [9], as shown in Fig. 9. As described in Sec. III A, kink A
stratification occurs over a wide range of flow rates, size
ratios, and system sizes for bidisperse spherical particles with
equal repose angles, which indicates a different mechanism
for stratification from those for mixtures whose constituent
particles have different repose angles [1113,18].
Careful observation of stratification for spherical particles
in our experiments reveals that the driving mechanism for the
FIG. 10. (Color online) Illustration of stratum formation at R =
3.4, W = 91 cm, and q = 0.9 cm2 /s. (a) and (b) show two stationary
Though Williams [3] and Baxter et al. [16] observed stratification states separated by a single avalanche. During the avalanche, kink A
with smooth spherical particles, Makse et al. [11,22] argued that some moves downslope and kink B forms. Dark (blue online): 1.69-mm
shape-induced stratification may have occurred. particles; light (red online): 0.5-mm particles.

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

forms a kink far from the bounding wall [kink A in Fig. 10(a)]. particle size ratio, silo width, and silo gap thickness. At
The kink has a larger local angle of repose than the slope small size ratios (e.g., R = 1.3), segregation in the flowing
of the heap, along with a layer of large particles on top of a layer is weak, which results in small concentration differences
layer of smaller particles behind it. The formation of the kink between strata. The silo also needs to be sufficiently wide so
on the slope is different from the formation of the kink for that a kink can form somewhere on the slope of the heap. For
mixtures of particles differing in both size and shape that is example, no stratification occurs for R = 1.5 and W = 46 cm,
initiated at the boundaries, as reported by Makse et al. [11] while stratification does occur for large widths (see Fig. 5).
and Gray and Hutter [12]. The kink formed on the slope in our Silo gap thickness also affects stratification as discussed in the
experiments is similar to the trapped kink on the slope of the Appendix.
heap reported by Shimokawa and Ohta [17,18] for large rough
and small smooth particles at very low flow rates (<O(1) g/s) B. Control parameters for segregation
and sufficiently large silo widths. However, as shown in Fig. 9,
since the angles of repose of each component of the mixtures The dynamics and mechanisms for segregation in heap flow
in our experiments are nearly the same, the mechanism for the have been previously studied [37] as discussed in Sec. I.
initiation of the kink on the slope for different-sized spherical At its simplest, the competition between percolation of two
particles is not the same as that proposed by Shimokawa and different-sized components perpendicular to the flow direction
Ohta [17,18]. and the advection of the mean flow determines the degree of
In our experiments, when kink A is formed, a shock wave final segregation. For instance, if percolation of small particles
similar to that in Ref. [12] propagates upslope and freezes downward through the flowing layer takes longer than the
the two layers (large particles above small particles) behind time to reach the downstream region of the heap, more small
the kink [Fig. 10(a)]. Since the local angle of repose at the particles will accumulate in the downstream region of the heap
kink is larger than the static angle of repose, the kink is along with large particles, resulting in a more mixed state. On
metastable. When a new avalanche occurs upstream on top the other hand, when small particles percolate quickly to the
of the frozen kink A avalanche [Fig. 10(b)], the kink A bottom of the flowing layer, they remain in the upstream region
avalanche restarts, resulting in the upstream stratified layers of the heap, resulting in stronger segregation.
moving further downstream. Depending on the width of the Based on this picture, we define a dimensionless time t =
silo and the feed rate, an avalanche may stop more than once tp /td , where tp represents the time scale for percolation, and
on the slope before it reaches the bounding wall or is overrun td represents the time scale for downstream convection by
by another kink. the mean flow. t  1 indicates better mixing, while t  1
Theoretical modeling of the stratification of different-sized indicates stronger segregation. Assuming a constant average
spherical particles depends on the physical mechanism for downstream velocity v d that decreases linearly with depth in
initiating kinks on the slope and incorporating this mechanism the flowing layer for simplicity (similar to previous work on
into a proper statistical avalanche model. Since the angle monodisperse heap flow [21,32]), we obtain td = L/v d , tp =
of repose for different components of the mixtures in our /v p , where v p is the mean percolation velocity and is the
experiments is nearly the same (Fig. 9), a mechanism that mean thickness of the flowing layer. Taking = vr L/v d based
depends on differences in the angle of repose between on mass conservation (i.e., q = vr L = v d ), the dimensionless
different components of the mixtures to induce stratification time scale can be expressed as a velocity ratio
[11,13,17,18] is not applicable to the formation of kinks in our
t = vr /v p . (1)
experiments. One possible alternative mechanism is associated
with a model proposed by Gray et al. [19,20], in which Equation (1) shows that for a particular percolation velocity
a segregation-mobility feedback and the deposition of large v p , which depends on the size ratio R, the degree of final
particles at the avalanche front can halt the flow. segregation depends only on vr . Increasing vr increases
Furthermore, due to the somewhat random nature of the t, indicating a decrease in segregation and an increase in
avalanches, the locations of the kinks formed on the slope mixing. For constant vr , t decreases as v p increases (e.g.,
are also somewhat random. Consequently, the lengths of the R increases). This means that at the same flow conditions,
stratified layers along the flow direction and widths of the mixtures with large size differences should have a higher
stratified layers normal to the flow direction vary (see Fig. 10), degree of segregation. Figure 11(a) shows profiles of L/L as
resulting in the nonperiodic stratified layers evident in our a function of vr for several size ratios (R = 1.5,2.2,3.4,4,6).
experiments. This differs from the results of Makse et al. As predicted by Eq. (1), L/L, a measure of segregation,
[11,13] and Gray and Hutter [12], where the kink is always decreases as vr increases at constant R. Furthermore, at
formed at the base or bounding walls of the silo, and the constant vr , as vp is increased by increasing R, L/L
stratified layers are nominally periodic. Therefore, statistical increases, indicating stronger segregation at larger R.
models of avalanches [2831] may also be useful for modeling To further examine the dependence of segregation on t
stratification. while varying vr and v p simultaneously, a relation between
Stratification disappears when no kink is formed on the the percolation velocity v p and experimental parameters such
slope. At higher flow rates, where surface flow is continuous, as R and strain rate at different q is needed. Bridgwater
large particles are continually advected to the downstream and colleagues [33,34] systemically studied the influence of
region of the heap, while small particles settle in the upstream different parameters, including the size ratio, density ratio,
region, resulting in the typical segregation pattern shown in strain rate, and normal stress, on percolation velocities in
Fig. 2(b). In addition, stratification is also limited by the various sheared systems. They found that particle size ratio


(a) 0.5 R investigated here. Further, as mentioned by Fan and Hill [26],
= 1.5 a shear gradient can also drive segregation in the spanwise
= 2.2 direction across the silo. At larger R, the shear effect is more
0.4 = 3.4
= 4.0
important than at smaller R. In our experiments, at R = 6 we
= 6.0 observed signs of both effects. Small particles preferentially fill
0.3 the voids between large particles at the sidewalls, which causes
a variation of particle concentration between the sidewalls.
0.2 Based on observations at the top of the free surface, large
particles also tended to gather toward the middle of the gap
away from the sidewalls for R = 6, probably due to the shear
0.1 effect [26].
At small size ratios (e.g., R = 1.3), no stratification was
0 observed because of the small size difference, as discussed
0 2 4 6 8
vr (cm/s)
in Sec. IV A. Segregation also becomes much weaker at
(b) 0.5 small size ratios and exhibits different characteristics than
at other size ratios. As mentioned by Goyal and Tomassone
[9], when R is smaller than a critical value of 1.4, the
0.4 concentration of each species changes continuously along
the heap, as shown for small q in Fig. 12(a) for R = 1.3.
For R > 1.4, segregation is more complete and boundaries
between the two segregated regions are sharper, as shown
in Fig. 7. Our results show that the transition between these
0.2 two different segregated states additionally depends on q and
W , as shown in Fig. 12. At W = 46 cm for R = 1.3 and
within the entire range of q in our experiments, continuously
varying segregation is always observed [Fig. 12(a)]. However,
for W = 91 cm, more complete segregation occurs at lower
0 q and continuously varying segregation occurs at higher q
0 0.5 1 [Fig. 12(b)]. One possible reason for this behavior is that at
t ~ vr s small R ordinary diffusion, which causes remixing of different
components [3537], may become comparable in importance
FIG. 11. (Color online) (a) L/L vs vr at different R from
to percolation over shorter W . Diffusion in flowing granular
1.5 to 6. At each R, data for different values of W use the same
materials is related to the flow kinematics such as local shear
symbols. (b) L/L as a function of the dimensionless time scale rates [3840], which are closely associated with the flow rate
t vr / g(R 1)Ds .
and system size. At small flow rates or large silo widths,
has the greatest influence on the percolation velocity, while diffusion could be weaker than percolation so that segregation
the others have little effect. However, an analytic relation is is stronger.
lacking. Here, to first order of approximation, we assume
v p depends only on R along with gravity, which drives the
100 (a)
flow. Figure
11(b) shows L/L as a function of t assuming small q
v p g(R 1)Ds , where R 1 is used to enforce zero 90
large q
percolation velocity for monodisperse particles. Curves of
L/L at different R collapse well compared with Fig. 11(a). 80
However, for R = 1.5, the scaling does not collapse the curves
for t > 0.07 (corresponding to vr > 0.5 cm/s), presumably 70
I (arbitrary units)

because in this situation, other mechanisms such as ordinary

diffusion also play important roles in segregation. Thus, the
expression we use for the percolation velocity is not always (b)
applicable. 100

C. Upper and lower limits of size ratio 80

In this work we considered bidisperse mixtures of spherical 60

particles with size ratios from 1.3 to 6.0, a relatively large range
compared to previous studies (Table I). We choose R  6 for 40
two reasons. First, as shown in Refs. 0 0.2 0.4 0.6 0.8 1
[2325] and references x/L
therein, when R is larger than (2/ 3 1)1 = 6.464, a small
particle can percolate through the smallest voids between FIG. 12. (Color online) Intensity as a function of x/L for R = 1.3
three large particles without external agitation. Thus, the at q = 1.8 cm2 /s (black diamond) and q = 21.6 cm2 /s (red circle)
mechanism for segregation is different from that at the smaller for (a) W = 46 cm and (b) W = 91 cm.

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

(a) 10 cm (b) (c) of the experiment and corresponding to the top of the vertical
band of particles, there are only larger particles close to the
end wall, indicating a strong decrease in bouncing-induced
segregation. In Figs. 13(b) and 13(c), in addition to the fact
that no smaller particles are observed in the downstream
regions of the silo next to the right wall, the boundaries
between the larger and smaller particle region are vertical
(essentially independent of the fall height), indicating that
(d) (e) bouncing-induced segregation has a negligible influence on
the final particle distributions.
Figures 13(d) and 13(e) compare segregation when the
granular mixtures have similar size ratios (R 2) but different
absolute sizes [2 mm and 1 mm in Fig. 13(d) and 1.1 mm and
0.5 mm in Fig. 13(e)]. The mixture with the larger absolute
size exhibits stronger bouncing-induced segregation [smaller
particles accumulate at the downstream wall of the silo in
Fig. 13(d)] than the mixture of smaller absolute size [no smaller
FIG. 13. (a)(c) Images showing reduction in bouncing-induced
segregation with increasing silo width W at R = 4 with 2-mm (black)
particles accumulate at the downstream wall of the silo in
and 0.5-mm (light) glass particles and q = 14 cm2 /s. (a) W = Fig. 13(e)]. Inertial and gravitational forces of the particles are
22 cm, (b) W = 46 cm, and (c) W = 69 cm. (d) and (e) Images proportional to the cube of particle radius, while air drag is
at similar size ratio (R 2) showing reduction in bouncing-induced at most proportional to the square of particle radius (at high
segregation with decreasing absolute particle size at W = 69 cm and Reynolds numbers) [41]. Therefore, the ratio of inertial or
q = 140 cm2 /s. (d) 2.0-mm (black) and 1.0-mm (light) particles. gravitational forces to air resistance is greater for the mixture
(e) 1.1-mm (black) and 0.5-mm (light) particles. of larger particles than for the mixture of smaller particles.
As a result, the smaller particles in the mixture bounce farther
after impact [Fig. 13(d)].
D. Bouncing-induced segregation This cursory study of bouncing-induced segregation
When the falling particles feeding the heap impact the demonstrates that bouncing-induced segregation can compete
top of the heap, they sometimes bounce toward the end of with free surface segregation and may result in different final
the silo. Since smaller particles generally gain momentum segregation patterns. However, as long as the silo width is large
when colliding with larger particles (similar to when a lighter enough, bouncing-induced segregation is minimized and free
golf ball collides with a heavier basketball), they tend to surface segregation dominates the final particle distributions.
bounce further down the heap than larger particles. This can To further study the effects of bouncing-induced segregation in
result in segregation that is reversed and unrelated to that heap flow, alternative methods to feed particles to the top of the
in the flowing layer. Bouncing-induced segregation, briefly heap (e.g., similar to that in Ref. [42]) may be needed to isolate
mentioned by Drahun and Bridgwater [5], causes smaller the effects of these two different segregation mechanisms.
particles to segregate to the downstream region of the heap,
which is, of course, opposite to what occurs during the free
surface segregation studied in this paper.
Several factors can influence bouncing-induced segregation In this paper we have shown that for mixtures of bidisperse
including silo width, fall height of the particle feed stream, spherical particles, three different final configurations in heap
and relative and absolute particle size. Figure 13 shows the flowstratified, segregated, and mixedcan be obtained by
influence of these factors on bouncing-induced segregation. In controlling flow properties and particle size ratios. Stratifica-
Figs. 13(a)13(c), W is varied while all other parameters are tion is associated with the formation of kinks on the slope of the
fixed. For the smallest silo width, W = 22 cm [Fig. 13(a)], a heap away from the bounding walls during discrete avalanches
narrow vertical band of small light-colored particles forms at at low flow rates (q smaller than 10 cm2 /s) at most size ratios
the downstream end of the heap (right wall). This is opposite (R > 1.4). The silo width W , or alternatively the flowing layer
to what is observed under normal free surface segregation length L, should be large enough that a kink can form along the
conditions. When W is increased to 46 and 69 cm [Figs. 13(b) slope of the heap (instead of at the end of the heap) during the
and 13(c)], no smaller particles are observed adjacent to avalanche. The transition between stratified and unstratified
the right end wall. Even though bouncing of particles still states is governed by q. When q is larger than a transitional
occurs at larger W , the bouncing particles reenter the flowing value dependent on R, the heap flow changes from a discrete
layer before reaching the end of the silo, so that bouncing- avalanche regime to a continuous flow regime with segregation
induced segregation is negligible. The effect of fall height in which neither kinks nor stratification are observed. The
on bounce-induced segregation can be seen by considering a degree of segregation is determined by competition between
single experimental run [see Fig. 13(a)]. As the heap grows, advection by the mean flow and percolation through the
fall height decreases, so that bouncing-induced segregation flowing layer. This competition can be characterized by the
becomes weaker. Consequently, in Fig. 13(a), the smaller ratio of the rise velocity of the heap vr to the percolation
particle region at the right wall of the silo becomes narrower velocity v p , which mainly depends on R. At the same R, as vr
from bottom to top. At the smallest fall height near the end increases, the degree of segregation decreases and eventually


transitions to a mixed state. The transitional rise velocity 0.5

becomes larger as R increases. T/D = 4.3

There are three major points, among others, that this study T/D = 4.3
raises: (i) Stratification of different-sized spherical particles is 0.4 T/D = 4.3
observed for a wide range of flow rates and size ratios, but T/Dl = 5.3
the dynamics of stratification appear different from those for T/D = 5.3
0.3 l
stratification of different size and shape particles observed in T/Dl = 5.3

previous research [1115,1720]. These apparent differences T/D = 6.4
in the physical mechanisms for stratification need further study, 0.2 T/D = 8.5
particularly the mechanism for initiating a kink on the slope
away from the bounding walls, which might be related to the
deposition of large particles at the avalanche front [19,20].
(ii) Mechanisms other than percolation play important roles in
heap segregation in the following situations: (a) At small size 0
0 2 4 6 8
ratios (R < 1.5), ordinary diffusion appears to become com- v r (cm /s)
parable to percolation, suggesting that a different segregation
configuration occurs (continuously varying segregation). (b) FIG. 14. L/L vs vr for mixtures of 2.98-mm and 0.5-mm
For R near the large size ratios at which spontaneous percola- particles (R = 6) at different silo gap thicknesses T and silo widths
tion can occur (R 6.5), wall effects or shear-induced segre- W , showing segregation is insensitive to T for T > 4Dl . Open
gation may cause spanwise segregation between the sidewalls. symbols: W = 91 cm; filled symbols: W = 69 cm (gray) and W =
Based on recent progress in modeling gravity-driven seg- 46 cm (dark).
regation [36,37,40,43] and shear-driven segregation [44,45]
in dense flows, developing a rigorous predictive segregation
model including mechanisms for the local particle distributions
in bounded heap flow is feasible with the proper kinematic and
segregation properties including the velocity field, stress par-
tition coefficients between small and large particles, and diffu- The effect of quasi-2D silo gap thickness T on stratification
sion coefficients. However, unlike other free surface flows such and segregation was examined for several binary mixtures at
as inclined chute flow and unbounded heap flow [27], bounded various q to determine whether T has a significant influence
heap flow has complicated kinematics in the continuous flow on the results. Silo thickness T was varied from 0.64 to
regime in the sense that both velocity components have 2.54 cm (corresponding to 3 < T /Dl < 12), and q was varied
gradients in both the streamwise and normal directions. This from 1.2 to 331 cm2 /s. Measurements of stratification and
suggests a detailed study of the kinematics of segregating flow segregation are carried out using the same criteria as those
in bounded heap flow is a necessary precursor to apply these in Sec. III.
models. (iii) When particles have sufficient impact velocity or The effects of T on segregation are shown in Fig. 14
the silo is not wide enough, bouncing of particles after heap by plotting L/L vs vr for R = 6. Within the range of T
impact can cause segregation opposite the usual segregation in investigated, the degree of segregation does not depend on
the flowing layer. Alternative methods of feeding particles to T , as long as T > 4Dl . The same trends are observed for
the top of the heap could potentially help isolate the effects of other values of R. In contrast, stratification is influenced by
bouncing-induced segregation from free surface segregation. T , but no clear trend is observed. For example, at R = 6,
We have only studied quasi-2D configurations of heap flow. when T increases from 1.27 to 2.54 cm, the stratification
While we expect that our results are indicative of those in 3D metric /I decreases at the same q. At T = 2.54 cm, /I
heaps, further work is needed to confirm this. Furthermore, we is relatively small over the entire range of q, indicating that at
have only considered the steady filling stage of heap formation this T , stratification is barely observable for all experimental
[see Fig. 1(b)], leaving initial heap formation and heap growth parameters. However, at R = 2 or R = 3.4, and certain values
for future investigation. of q, /I at larger T is larger than at smaller T , indicating that
stratification does not always monotonically decrease when
T increases at all R. The mechanism for the dependence
of stratification on T remains unclear and needs further
We are grateful for the laboratory assistance of Emre Yildiz investigation. In this paper, we use T = 1.27 cm for all
and helpful discussions with Karl Jacob and Ben Freireich. We experiments, as it is sufficient for achieving T -independent
also acknowledge financial support from The Dow Chemical segregation and it minimizes the volume of particles needed
Company. to perform the experiments.

[1] P. Evesque and J. Rajchenbach, Phys. Rev. Lett. 62, 45 (1988). [5] J. A. Drahun and J. Bridgwater, Powder Technol. 36, 39
[2] P. G. de Gennes, Rev. Mod. Phys. 71, S374 (1999). (1983).
[3] J. C. Williams, Sheffield Univ., Fuel Soc., J. 14, 29 (1963). [6] K. Shinohara, K. Shoji, and T. Tanaka, Ind. Eng. Chem. Process
[4] J. C. Williams, Powder Technol. 2, 13 (1968). Des. Dev. 11, 369 (1972).

YI FAN et al. PHYSICAL REVIEW E 86, 051305 (2012)

[7] K. Shinohara and G. Enstad, Proceedings of the Second World [28] V. Frette, K. Christensen, A. Malthe-Sorenssen, J. Feder,
Congress Particle Technology (Society of Powder Technology, T. Jossang, and P. Meakin, Nature (London) 379, 49 (1996).
Kyoto, Japan, 1990), p. 45. [29] C. M. Aegerter, R. Gunther, and R. J. Wijngaarden, Phys. Rev.
[8] N. Thomas, Phys. Rev. E 62, 961 (2000). E 67, 051306 (2003).
[9] R. K. Goyal and M. S. Tomassone, Phys. Rev. E 74, 051301 [30] A. R. Abate, H. Katsuragi, and D. J. Durian, Phys. Rev. E 76,
(2006). 061301 (2007).
[10] M. Rahman, K. Shinohara, H. Zhu, A. Yu, and P. Zulli, Chem. [31] K. A. Dahmen, Y. Ben-Zion, and J. T. Uhl, Nat. Phys. 7, 554
Eng. Sci. 66, 6089 (2011). (2011).
[11] H. A. Makse, S. Havlin, P. R. King, and H. E. Stanley, Nature [32] D. V. Khakhar, A. V. Orpe, P. Andersen, and J. M. Ottino,
(London) 386, 379 (1997). J. Fluid Mech. 441, 255 (2001).
[12] J. M. N. T. Gray and K. Hutter, Continuum Mech. Thermodyn. [33] M. H. Cooke, J. Bridgwater, and A. M. Scott, Powder Technol.
9, 341 (1997). 21, 183 (1978).
[13] H. A. Makse, P. Cizeau, and H. E. Stanley, Phys. Rev. Lett. 78, [34] J. Bridgwater, M. H. Cooke, and A. M. Scott, Trans. Inst. Chem.
3298 (1997). Eng. 56, 157 (1978).
[14] Y. Grasselli and H. J. Herrmann, Granular Matter 1, 43 (1998). [35] K. M. Hill, D. V. Khakhar, J. F. Gilchrist, J. J. McCarthy, and
[15] J. P. Koeppe, M. Enz, and J. Kakalios, Phys. Rev. E 58, R4104 J. M. Ottino, Proc. Nat. Acad. Sci. USA 96, 11701 (1999).
(1998). [36] J. M. N. T. Gray and V. A. Chugunov, J. Fluid Mech. 569, 365
[16] J. Baxter, U. Tuzun, D. Heyes, I. Hayati, and P. Fredlund, Nature (2006).
(London) 391, 136 (1998). [37] S. Wiederseiner, N. Andreini, G. Epely-Chauvin, G. Moser,
[17] M. Shimokawa and S. Ohta, Phys. Lett. A 366, 591 (2007). M. Monnereau, J. M. N. T. Gray, and C. Ancey, Phys. Fluids 23,
[18] M. Shimokawa and S. Ohta, Phys. Rev. E 77, 011305 (2008). 013301 (2011).
[19] J. M. N. T. Gray and C. Ancey, J. Fluid Mech. 629, 387 [38] V. V. R. Natarajan, M. L. Hunt, and E. D. Taylor, J. Fluid Mech.
(2009). 304, 1 (1995).
[20] J. M. N. T. Gray and B. P. Kokelaar, J. Fluid Mech. 652, 105 [39] B. Utter and R. P. Behringer, Phys. Rev. E 69, 031308 (2004).
(2010). [40] B. Marks, P. Rognon, and I. Einav, J. Fluid Mech. 690, 499
[21] T. Boutreux and P.-G. de Gennes, J. Phys. I 6, 1295 (1996). (2011).
[22] P. Cizeau, H. A. Makse, and H. E. Stanley, Phys. Rev. E 59, [41] G. K. Batchelor, An Introduction to Fluid Dynamics (Cambridge
4408 (1999). University Press, Cambridge, UK, 2000).
[23] A. M. Scott and J. Bridgwater, Ind. Eng. Chem. Fundamen. 14, [42] A. Samadani and A. Kudrolli, Phys. Rev. E 64, 051301 (2001).
22 (1975). [43] J. M. N. T. Gray and A. R. Thornton, Proc. R. Soc. A 461, 1447
[24] S. B. Savage and C. K. K. Lun, J. Fluid Mech. 189, 311 (1988). (2005).
[25] F. Lomine and L. Oger, Phys. Rev. E 79, 051307 (2009). [44] L. B. H. May, L. A. Golick, K. C. Phillips, M. Shearer, and
[26] Y. Fan and K. M. Hill, Phys. Rev. Lett. 106, 218301 (2011). K. E. Daniels, Phys. Rev. E 81, 051301 (2010).
[27] GDR MiDi, Eur. Phys. J. E 14, 341 (2004). [45] Y. Fan and K. M. Hill, New J. Phys. 13, 095009 (2011).